Skip to main content
An SLO turns “this service should be reliable” into a target you can measure and alert on. You pick an indicator (an SLI) such as availability or p99 latency, set a target such as 99.9% over 30 days, and SRE Agent tracks the remaining error budget.
SLOs page with a Needs attention banner, status counts, and cards for payments availability and p99 latency showing compliance, budget left and burn rate

The SLOs page shows compliance, error budget and burn rate for each objective, with the ones that need attention at the top.

​
Create SLOs with the wizard

The wizard needs the Pro plan. It finds your workloads, proposes indicators, and checks each one against 30 days of your real data before you commit to a target.
1

Open the wizard

Go to SLOs and click New SLO (it opens the wizard), or open Manage SLIs and click SLO Wizard. From an alert or an investigation that carries a service label, click Generate SLO candidates → to skip straight to the probe step.
2

Pick a data source

Choose where the metrics live. The wizard lists Prometheus, CloudWatch Metrics and Loki sources. Click the one you want. If none appear, add one under Settings on the Data Sources tab.
3

Pick a workload

The table shows each workload’s name, Alerts (7d), Existing SLIs and Series. The noisiest workloads come first. Click Generate SLO candidates → on the one you want.
4

Review the candidates

The wizard proposes up to four indicators using the RED method: rate, errors and duration, plus saturation when CPU, memory or connection metrics exist. Dimensions with no matching metrics are left out. Results stream in, so you can act on one card while others are still probing.
5

Create the SLI and SLO

Each card shows the query, Observed 30d compliance and a Target. Click Create SLI + SLO on the cards you want. The card changes to Created. with a View SLI → link.
The recommended target sits about 0.2 percentage points below what the service actually achieved, never lower than 90%, so it is a target you can meet today and still tighten later. When there is no history to measure, the card says No historical data, using industry default and uses a typical target for that indicator type instead.
Members and admins can create SLOs. Viewers can read them.

​
Create an SLO on a Free plan

SLOs you define yourself are open on every plan, and so are tracking and alerting on them. Open SLOs > Manage SLIs and click Generate AI Suggestions to get SLI proposals. Click Approve and then Activate on the one you want, then click Add SLO on it. The Create SLO form asks for a Name, a Target (%) and a Window (days), defaulting to 30, plus the burn-rate settings listed in the SLO reference.

​
Create an SLO from a synthetic check

If you already probe an endpoint (synthetic checks need Pro), open the check under Synthetic Checks and click Create availability SLO or Create latency SLO. The SLO is built from that check’s own results. See Synthetic checks.

​
Read the dashboard

SLOs groups objectives by service. The cards at the top filter by Healthy, Warning, Breached, Degraded and Suspended, and Needs attention lists everything that is breached, degraded, warning, failing to compute or not yet computed. Click an SLO to see its compliance, remaining error budget, burn rates and the SLI and data source behind it. Recompute Now refreshes it on demand.

​
Burn and budget alerts

Open an SLO and use the pencil icon to edit it. Alert on burn rate, Alert on budget warning and Alert on tracking failure control when it alerts, and Gate deploys on this SLO controls its deploy gate. Breach alerts arrive like any other alert, so they can page people and trigger runbooks. The thresholds and defaults are in the SLO reference.

​
Degraded and no data

An SLO shows Degraded when SRE Agent ran the computation and does not trust the answer, for example when the data source is unreachable or returned nothing at all. The reason appears on the card. An empty answer is treated differently from a zero:
  • A service that served no requests counts as 100% compliant. Nothing violated the objective.
  • A query that returns no series, rows or datapoints is not reported as 100%. The SLO goes degraded, because a perfect month read off broken monitoring would be misleading.
A degraded SLO recovers by itself: SRE Agent keeps retrying, and the first window with data clears the status and resolves the tracking alert. While it is degraded, it holds the service’s deploy gate. If an SLI shows Unverified or no data, open SLOs > Manage SLIs and click Re-probe once your metrics are flowing. To get good SLIs in the first place, emit one request counter and one latency histogram per service, labelled with the service name, HTTP method, route template and status code.

​
Suspend and resume

Click Suspend on an SLO (any member can) to pause its computation and alerting, and Resume to restart it. The More menu on the dashboard has Maintenance mode (suspend all) and Resume all suspended. Suspending an SLO resolves its own open alerts. A suspended SLO never blocks a deploy, but the gate still reports it so nobody mistakes “not watched” for “healthy”.